• Lvxferre
    link
    fedilink
    English
    arrow-up
    2
    arrow-down
    1
    ·
    edit-2
    16 hours ago

    Why are we using LLMs as calculators?

    Two reasons: // 1. Humans always try to use any new software/hardware we invent to do calculation // 2. We don’t actually want them to do math for the sake of replacing calculators, we want to understand if they can reason their way to AGI.

    And the author fails to mention the most important reason:

    Maths is so universal that the ability to perform simple maths is required for any decent output, regardless of “FEEL THE AGI! IT IS COMING!!!” shite or human tendencies.

    And it doesn’t take a bloody genius to realise that.

    I’ll give you guys an example. The author themself listed LLMs as able to write poems… except that they can’t even get the number of syllables right, because guess what, they suck at counting! Like this:

    The prompt is “Please, write a sonnet in Portuguese about margarine pots. Use verses with ten metric syllables each, and ABBA CDDC EFE GFG rhymes.” The first line already fucks it up - it has 11 metric syllables, not 10.

    (The rhymes are also fucked up. I won’t judge it on semantic grounds because the prompt was silly, but “tradition”? Margarine??? Pffft.)

    But WhY ArE We UsInG LlMs As CaLcUlAtOrZ, right?

    However, evaluating how good LLMs are at calculation doesn’t take into account a critical component: the way that calculators arrive at their answer is radically different from how these models work.

    It doesn’t take into account your “ackshyually” because your “ackshyually” is a bloody red herring.

    I’ve stopped reading this idiocy here, as I’ve already wasted too much time with it.